The short version
A CPU follows instructions, billions of them every second. Every program, from a game to a web browser, is broken down into tiny steps like "add these two numbers" or "copy this from memory". How fast a CPU feels comes down to four things:
- Clock speedHow many times per second each core ticks (5.5 GHz = 5.5 billion ticks).
- Work per tickHow many instructions a core gets through each tick. Newer designs do more, so GHz alone doesn't tell you speed.
- CoresHow many jobs it can run at once. Great for video and rendering; games only use a handful.
- CacheFast memory on the chip. Big caches are why X3D chips top the gaming charts.
The basic loop: fetch, decode, execute
What one core does, over and overA program sits in memory as a list of instructions. The core keeps a program counter pointing at the next one, and a few tiny, ultra-fast storage slots called registers to work in. Watch it run a four-instruction game loop that adds 10 to the score.
Clock speed: how fast the core ticks
Every step above happens on the tick of a clock. A core boosting to 5.5 GHz ticks 5.5 billion times a second, so each tick lasts about 0.18 nanoseconds. In that time, light travels only about 5.5 cm (just over 2 inches).
- Base vs boost. The base clock is what the chip promises under full load; the boost clock is the most it reaches, usually on one or two cores at a time. How long it holds boost depends on your cooler and motherboard.
- GHz only compares chips of the same design. A newer core that does more work per tick can beat an older one with a higher clock. That's why benchmarks matter more than the number on the box.
The assembly line: pipelining and guessing
Simplified to 4 stages; real cores have 15–20Running one instruction start to finish before touching the next would leave most of the core idle. Instead, cores work like an assembly line. Switch between the three modes below.
- Real cores go much further. They run several instructions in the same tick and reorder them to dodge waits, as long as the result comes out the same.
- Guessing is a big deal. Programs are full of if/then branches. Modern branch predictors guess right well over 90% of the time, but each wrong guess wastes a dozen or more ticks on a real, longer pipeline.
- Waiting on memory is the other big stall. If an instruction needs data that isn't in cache, it can wait hundreds of ticks for RAM (see the speed ladder in How RAM works).
Cores: doing several things at once
Made-up numbers to show the idea; real games and apps varyEach core is a complete copy of everything above, so an 8-core CPU can run 8 instruction streams at once. But work only speeds up if it can be split. Pick a workload and a number of cores.
- Threads. AMD Ryzen cores can each juggle two instruction streams (simultaneous multithreading), filling gaps when one stream stalls. That's where "8 cores / 16 threads" comes from. It adds roughly 10–30% on heavy multi-threaded work, not double.
- Performance and efficient cores. Intel Core Ultra chips mix big performance cores (P-cores) with smaller efficient cores (E-cores). Games and other demanding single jobs run on the P-cores; the E-cores add a lot of muscle for multi-threaded work.
Cache, and why X3D chips win at games
Cache is small, very fast memory built into the CPU. When the data a core needs is already in cache, it arrives in a few nanoseconds; when it isn't, the core waits around 75 ns for RAM, which is hundreds of ticks. Games constantly jump around a large amount of data, so they miss the cache a lot.
AMD's X3D chips stack an extra 64MB layer of cache onto the processor, tripling the 32MB L3 cache of an 8-core Ryzen to 96MB. More of the game fits on the chip, the core waits on RAM less often, and frame rates rise. Most other apps don't benefit nearly as much.
On 16-core chips like the Ryzen 9 9950X3D, only one of its two 8-core halves gets the extra cache; Windows and AMD's driver steer games onto that half.
What the benchmarks show
Every CPU we track, fewest cores firstCompare the two columns. In games, the 8-core X3D chips sit at the top, ahead of chips with two or three times as many cores. In multi-threaded apps like rendering and video exports, core count takes over.
Power and heat
Every tick switches billions of transistors, and that turns into heat. Chips boost as high as their power limit and temperature allow, so a better cooler can mean higher sustained speeds. Power use varies a lot: a Ryzen 5 9600X is rated at 88 W maximum, while a Core Ultra 9 285K can draw up to 250 W. That decides the cooler you need, and it adds up in the power supply calculator.
What this means when you buy
- Mainly gaming: an 8-core X3D chip is the fastest choice; past 8 cores, games barely improve. Pair it with the best graphics card you can afford, since that usually limits frame rate more (see Which graphics card?).
- Gaming on a budget: a 6-core chip like the Ryzen 5 9600X still runs every current game well and leaves money for the graphics card.
- Video editing, 3D rendering, compiling code: more cores pay off directly. Look at 12–24 core chips and the multi-threaded benchmarks.
- Pick the CPU, then the board. The CPU's socket decides the motherboard (see the motherboard chipset guide).
Today: Ryzen 7 9800X3D from …, Ryzen 5 9600X from …, Ryzen 9 9950X3D from ….